See what
you're making room for.

A place for your next idea

A good idea has terrible timing. It shows up while your desk is already full, and disappears when you finally find somewhere to put it. PHNTM One gives that thought a place to land.

Your desk. Your device.

This is a dedicated assistant built around a Raspberry Pi five, eight gigabytes of RAM, and a ten point one inch touchscreen. Local chat, notes, remembered details and document questions, in one place.

Start messy. Keep the original.

Let’s use a fictional workshop project: a small shelf for a crowded desk. Write the rough thought in Notes. The original is saved before the model starts working. Dictation can capture up to sixty seconds, with a transcript to review.

A draft you can inspect.

Ask for a project checkpoint or a cleaner draft. Your original stays separate. A model should help you develop an idea. It should not quietly replace what you actually said. Review the result before trusting it.

Ask about your own reference.

Bring a reference document in through Phantom Drop on your local network. It is read and indexed on this device. Then ask a focused question. File and page references give you somewhere to check the answer. OCR and models can still make mistakes.

A conversation with somewhere to go.

Chat runs on the small local model in Private mode. It can help you think through a task, draft a response, or ask about something you supplied. It is not a frontier cloud model. Answers take time. This film trims waiting; it is not a speed benchmark.

Leave tomorrow a breadcrumb.

Explicitly save a useful detail as a memory, so the next conversation has context. The Library lets you inspect stored memories and documents. The constellation view connects memories that share keywords.

Make the next step visible.

Tasks and reminders give the next step somewhere to live. They depend on a trustworthy clock. After offline power loss, time may need verification. A useful assistant should show uncertainty rather than dress it up as confidence.

Keep the boundary clear.

In Private mode, the model, memory search, transcription and speech run on the device. A paired phone browser communicates across your local network. Optional Claude answers send context to the provider. Connected Telegram, email and calendar have separate data paths. Local inference does not turn those services offline.

The engineering under the glow.

The orb looks nice. It has never once solved a RAM problem. The model, audio, interface and document workers share limited resources. The design reserves headroom and releases idle models. Reloading one can make the next answer slower. Response time and reliability work continues. These are real tradeoffs, not disappearing fine print.

A dedicated place, built by a person.

PHNTM One is built by Jacob DeCamp at DeCampLabs. It is a commercial product, made to order for seven hundred ninety nine dollars. No required subscription for local use. Optional cloud services may have their own charges. See the owner manual, specifications and build story at phantom core dot com. Bring your next idea.

No spoken narration; quiet instrumental music. The recording begins with a PHNTM One title, then a hand operates the physical touchscreen through Device health, Memory, Notes, and Reminders. The typing section is labeled and accelerated to three times its original speed. A saved reminder is shown, followed by Chat and the Home command center. The closing title reads “Meet PHNTM One” and “www.phntmcore.com”.

The footage is edited for length and clarity. It does not measure AI response speed. On-screen readings describe the filmed unit at that moment.

A cinematic sequence shows the touchscreen device, a person at a workstation, close-ups of an illustrated interface, and green light effects. Created with AI from product photographs.